Salesforce UI Automation: POM vs UTAM vs TestZeus Hercules

Reading Time: 7 minutes

Having joined my first Salesforce project just over six months ago, we’re now ready to consider automating the UI. We previously had one automated UI flow for our old system, which was completely different, and this was created as more of a proof of concept (PoC) for automated system integration tests across different products. The main thing I learned while maintaining this old automation is that Salesforce randomly changes things, such that UI automation breaks for no good reason. I’m talking different locators, altered HTML structure, and removing auto-load behaviours. It’s pretty annoying, to say the least.

So this time round, we’re taking all the learnings up until now and creating new automation infrastructure from scratch. I want to share a little bit about which approach we chose, and why. I struggled to find resources myself, so I hope this will be helpful to others. I’m also currently going through the Foundation Certificate in Test Automation from Ministry of Testing, which has a section on tool selection. The timing is a happy coincidence, and I thought it would be nice to share a real example of tool evaluation and selection.

Additional Context

As additional context for this evaluation:

I’m the first and only Quality Engineer on the project
Although I’ve written some automation in the past, it’s a skill I want to develop further
Like many teams, we have many requirements and limited resources; it’s very important that the benefits we get are worth the investment we make
We agreed that automating some key flows could be useful, but we do not plan to have hundreds of UI automation scripts
Our existing system integration tests use Playwright and TypeScript
We have a company Tech Radar, and would like to use tools already on it, where possible

Criteria

Naturally, there are some criteria which our chosen solution should meet:

Easy to set up
As little maintenance effort as possible
Simple to learn
Good resources / documentation / community
Likely to last / unlikely to be discontinued
Free to use
“Approved” for company use

Potential Approaches

There are three main approaches that I considered for our Salesforce UI automation:

The classic page object model (POM) pattern
1. A popular approach, which makes logical sense to my brain and reduces repetition of locator definitions
Salesforce’s own UI test automation model (UTAM) framework
1. An option which specifically claims to reduce the pain of maintaining locators
TestZeus’ open source testing agent, Hercules
1. A technology with self-healing AI features, which theoretically means total removal of locator maintenance

Comparison

	POM	UTAM	Hercules
Easy to set up	No additional tools or frameworks required – our existing tech stack (Playwright + TypeScript) can be used More control over locators Initial set-up of page objects can be time consuming	Built specifically for Salesforce and handling shadow DOMs Standard components are included Requires third-party web-driver UTAM files are written in JSON, then compiled to page objects Custom components need to be manually set up Less control over locators	Supports automation of Salesforce Video guides for set up Requires knowledge of Python and Gherkin Free version requires own AI agent, which may not be as effective
As little maintenance effort as possible	Locators only need to be updated in one place Locators need to be updated manually Generally, no good strategy for Salesforce locators exists	Locators for standard components are updated with each new UTAM version UTAM is sometimes updated after Salesforce, so automation still breaks Custom components still need to be manually maintained	Self-healing locators require no maintenance
Simple for the team to learn	Simple, logical structure Requires knowledge of Playwright and TypeScript	Requires knowledge of JSON, not TypeScript Requires knowledge of UTAM grammar	No code required for writing automation Requires knowledge of Gherkin
Good resources / documentation / community	Widely used and documented	Tutorials aren’t based on Salesforce apps Not many resources or much engagement	Video guides Written documentation is mostly for initial set up Still fairly new
Likely to last / unlikely to be discontinued	Established pattern without third-party dependencies	Hasn’t been updated in a while Not a focus for Salesforce	Creators are passionate about having an open-source option Still fairly new
Free to use	Free	Free	Hercules is free AI agent may incur additional costs
“Approved” for company use	Technologies on Tech Radar	Not yet on Tech Radar	Not yet on Tech Radar

Pros and Cons

POM + Playwright + TypeScript

Pros	Cons
Adopted company technologies	Flaky locators
Widely used pattern	No good locator strategy
Built-in driver
Separation of concerns
Full control over locators

UTAM + JSON + WebDriver

Pros	Cons
Built to handle shadow DOMs	New to company
JSON for standard components updated automatically	No built-in driver
	UTAM files need to be manually regenerated when JSON changes
	JSON for custom components still requires manual maintenance
	More logic in automation scripts
	Less flexibility for locators

Hercules + Playwright + Python + Gherkin

Pros	Cons
Al agent	New to company
No code required for writing automation	Free tier doesn’t include specialised AI agents
Self-healing locators	AI agent may incur additional costs
Open-source version still supports Salesforce	Little opportunity for developing programming skills

Recommendation

Based on my research and evaluation, my recommendation for our project is to use POM.

Although POM does not provide us with Salesforce-specific capabilities, it does allow us full control over locators, so we have flexibility in our problem solving.

While UTAM is specifically built for Salesforce, locators still need to be maintained, and I don’t have confidence in it’s longevity. As soon as it stops being maintained, we’ll lose the benefits and cannot reuse anything.

Hercules has the most appeal, from a maintenance standpoint, but uncertainties around the effectiveness and cost of the AI agent, coupled with the other cons, make it less attractive for our context.

I hope you found this helpful. If you’re in a similar situation, I’d love to hear from you. What approach(es) have you tried, and what challenges have you faced? Let me know in the comments.

Disclosure: As part of my research, I spoke with Robin Gupta, the founder of TestZeus, to get his opinion on UTAM. However, he was very careful not to make the conversation about TestZeus, and we only spoke about it when I specifically asked him for information on it.

Discover more from Cassandra HL

Subscribe to get the latest posts sent to your email.